sensing·
⏱ 45s
TINY sees through the head camera and can inject what it sees straight into the conversation. All sensing is read-only.
reachy_camera — grab a frame·
Captures a single frame from the head camera via the daemon. With no
save_path, it uses TINY_CAMERA_SNAPSHOT (default /tmp/tiny_view.jpg).
reachy_look_at — visual servoing·
Points the head so pixel (u,v) moves toward center. Chain it after a
detection to track a face or object.
take_photo — bidi vision (the good one)·
Grabs a frame and injects it into the model's context with your question — so TINY actually reasons about what it sees and replies aloud. This is the tool to use for "look at me", "what do you see?", "who's there?".
"look at me" → take_photo(question="who is in front of me?")
"what's this?" → take_photo(question="what object am I looking at?")
"read this to me" → take_photo(question="transcribe any text you see")
Media backend
The daemon owns the camera. If frame grabs fight the daemon, set
REACHY_MEDIA_BACKEND=no_media and let reachy_camera grab transiently, or
REACHY_CAMERA_BACKEND=local. See env vars.